The neuronal K-Cl cotransporter KCC2 influences postsynaptic AMPA receptor content and lateral

diffusion in dendritic spines

Grégory Gauvaina,b,c, Ingrid Chammaa,b,c,1, Quentin Chevya,b,c,1, Carolina Cabezasa,b,c, Theano Irinopouloua,b,c, Natalia Bodrug^a,b,c, Michèle Carnaud^a,b,c, Sabine Lévi^a,b,c,2, and Jean Christophe Poncer^a,b,c,2

aInstitut National de la Santé et de la Recherche Médicale, Unité Mixte de Recherche-S 839, F75005 Paris, France;bUniversité Pierre et Marie Curie, F75005 Paris, France; andcInstitut du Fer à Moulin, F75005 Paris, France

Edited* by Roger A. Nicoll, University of California, San Francisco, CA, and approved August 9, 2011 (received for review May 25, 2011)

The K-Cl cotransporter KCC2 plays an essential role in neuronal chloride homeostasis, and thereby influences the efficacy and po-larity of GABA signaling. Although KCC2 is expressed throughout the somatodendritic membrane, it is remarkably enriched in den-dritic spines, which host most glutamatergic synapses in cortical neurons. KCC2 has been shown to influence spine morphogenesis and functional maturation in developing neurons, but its function in mature dendritic spines remains unknown. Here, we report that suppressing KCC2 expression decreases the efficacy of excitatory synapses in mature hippocampal neurons. This effect correlates with a reduced postsynaptic aggregation of GluR1-containing AMPA receptors and is mimicked by a dominant negative mutant of KCC2 interaction with cytoskeleton but not by pharmacological suppression of KCC2 function. Single-particle tracking experiments reveal that suppressing KCC2 increases lateral diffusion of the mo-bile fraction of AMPA receptor subunit GluR1 in spines but not in adjacent dendritic shafts. Increased diffusion was also observed for transmembrane but not membrane-anchored recombinant neuro-nal cell adhesion molecules. We suggest that KCC2, likely through interactions with the actin cytoskeleton, hinders transmembrane protein diffusion, and thereby contributes to their confinement within dendritic spines.

T

Although KCC2 function primarily influences the efficacy of GABAergic signaling, its presence in dendritic spines (10) raises the question of its role in spine morphogenesis and function. Genetic ablation of KCC2 in mice compromises spine matura-tion and excitatory synapse formamatura-tion in immature hippocampal neurons (11). This effect appears to be independent of KCC2 function but, instead, involves KCC2 interaction with the neu-ronal FERM-domain protein 4.1N (12). However, KCC2 ex-pression is up-regulated during postnatal development and is maximal in mature neurons (13), after spine formation, where its role in the maintenance and function of dendritic spines remains unknown. Here, we show that suppression of KCC2 after spine morphogenesis reduces postsynaptic glutamate receptor content and relieves a constraint to the lateral diffusion and aggregation of these receptors and other transmembrane proteins within dendritic spines. This effect likely involves KCC2 interaction with submembrane actin cytoskeleton through its CTD but not

its ion transport function. Thus, suppression of KCC2 in mature neurons influences synaptic efficacy at glutamatergic synapses independent of GABAergic function.

Results

Suppression of KCC2 Expression in Mature Hippocampal Neurons.

KCC2 is expressed throughout the somatodendritic membrane of cortical neurons but also in dendritic spines (10). In mature [>28 days in vitro (DIV)] hippocampal neurons, anti-KCC2 immunostaining revealed numerous KCC2-immunopositive clu-sters within both spine heads and spine necks (Fig. 1 A and E). The intensity of KCC2 cluster immunostaining was 76% higher in dendritic spines than on dendritic shafts (P < 0.001), sug-gesting that KCC2 primarily aggregates in dendritic spines. We evaluated the role of KCC2 in dendritic spine maintenance and function using RNAi. In cultured hippocampal neurons, excit-atory synapses are formed from 5 to 7 DIV, whereas mature synapses onto dendritic spines appear from 8 to 10 DIV (14). Neurons were thus transfected at 14 DIV with plasmids ex-pressing GFP and either nontarget shRNA or shRNA against KCC2 and were processed 10 d later. The efficacy and specificity of RNAi on KCC2 expression were established at the protein level (Fig. 1 C–E) by comparing KCC2 and MAP2 immunore-activity in neurons expressing either construct. We selected one of four shRNA sequences leading to maximal reduction of KCC2 immunoreactivity (−86.4 ± 4.1% of control; P < 0.005). Over-expression of this sequence had no significant effect on MAP2 immunoreactivity (P = 0.2).

KCC2 suppression in mature neurons had no significant effect on the mean density (P = 0.1) or length (P = 0.2) of dendritic spines but caused a 30% increase in spine head diameter (P < 0.001) and an increase in the proportion of mushroom-type spines (Fig. 1 F and G andTable S1). This effect contrasts with the genetic ablation of KCC2, which prevents dendritic spine morphogenesis in immature neurons, leading to predominant long filopodia-like protrusions (11). We asked whether this discrepancy was related to the mode or the timing of KCC2 suppression. In hippocampal neurons transfected at 4 DIV, before spine forma-tion, suppression of KCC2 resulted in a 12% increase in spine length (P < 0.005) and a modest but significant reduction in spine head diameter (P < 0.05;Fig. S1). The overall spine density was unchanged (P = 0.9), but the proportion of filopodia-like pro-trusions was increased in neurons expressing shRNA against KCC2 (P < 0.02; Table S1). Therefore, suppression of KCC2 expression has opposite effects on spine morphology in mature vs.

Author contributions: S.L. and J.C.P. designed research; G.G., I.C., Q.C., C.C., N.B., S.L., and J.C.P. performed research; M.C. contributed new reagents/analytic tools; G.G., I.C., Q.C., T.I., S.L., and J.C.P. analyzed data; and S.L. and J.C.P. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor. 1I.C. and Q.C. contributed equally to this work.

2To whom correspondence may be addressed. E-mail: sabine.levi@inserm.fr or jean-christophe.poncer@inserm.fr.

This article contains supporting information online atwww.pnas.org/lookup/suppl/doi:10. 1073/pnas.1107893108/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1107893108 PNAS Early Edition | 1 of 6

N E U R O S C IE N C E

immature hippocampal neurons and KCC2 is not required for the anatomical maintenance of mature dendritic spines.

Reduced Quantal Size and GluR1 Accumulation in Dendritic Spines After KCC2 Suppression. Dendritic spine morphology has been correlated with synaptic function, particularly with postsynaptic density (PSD) size (15), postsynaptic receptor content (16), and lateral diffusion (17). Therefore, increased spine head volume upon KCC2 suppression might be expected to correlate with an increased number of postsynaptic receptors and synaptic strength (18). We tested this hypothesis by recording miniature excitatory postsynaptic currents (mEPSCs) from hippocampal neurons expressing nontarget or KCC2-specific shRNAs (Fig. 2 Aand B). Surprisingly, mEPSC amplitude was not increased but rather reduced in neurons expressing shRNA against KCC2 compared with nontarget shRNA (12.4 ± 0.9 pA vs. 14.9 ± 0.9 pA; P < 0.05). Suppression of KCC2, on the other hand, had no significant effect on mEPSC frequency (14.8 ± 1.9 Hz vs. 18.3 ± 2.3 Hz; P = 0.2), suggesting that the mean number of functional synapses was unaffected (Fig. 2B).

Spine enlargement may increase PSD size, and thus reduce the probability of postsynaptic receptor activation by released glu-tamate (19). However, we detected no difference in the onset kinetics of mEPSCs in neurons expressing KCC2-specific vs. nontarget shRNA (0.97 ± 0.03 ms vs. 0.95 ± 0.03 ms; P = 0.8; Fig. 2A). We thus asked whether suppressing KCC2 affected post-synaptic receptor density, by comparing GluR1 expression in neurons expressing shRNA against KCC2 compared with

non-target shRNA (Fig. 2 C and D). GluR1 immunostaining showed marked punctae primarily on dendritic spines and, to some ex-tent, on dendritic shafts (Fig. 2C). The relative intensity of GluR1 clusters within dendritic spines was significantly reduced in neu-rons expressing KCC2-specific shRNA (−46.9 ± 8.7%; P < 0.005; Fig. 2D). In contrast, the proportion of dendritic spines bearing GluR1 clusters was unaffected (83.1 ± 3.2% vs. 88.4 ± 2.1%, respectively; P = 0.2), consistent with the lack of effect of KCC2 suppression on mEPSC frequency. These results suggest that the reduction in EPSC quantal size induced by KCC2 suppression reflects a reduced density of postsynaptic AMPA receptors at excitatory synapses.

Preventing Molecular Interactions of KCC2 with Intracellular Partners Mimics KCC2 Suppression.KCC2 suppression may increase chlo-ride concentration (20) and thereby reduce GABAergic signal-ing. This may subsequently lead to scaling of excitatory synapses through homeostatic plasticity (21). We tested this hypothesis using chronic application of the KCC2-specific antagonist VU0240551 (22). Treatment of 21 DIV neurons with 6 μM VU0240551 in DMSO for longer than 72 h induced a significant increase in spine head volume compared with treatment with DMSO alone (P < 0.001; Table S1). However, no significant change was observed in the mean amplitude (18.7 ± 1.1 pA vs. 16.8 ± 0.8 pA; P = 0.2), frequency (30.1 ± 3.1 Hz vs. 31.3 ± 3.0 Hz; P = 1.0), or onset kinetics (1.03 ± 0.02 ms vs. 1.07 ± 0.02 ms; Fig. 1. Suppression of KCC2 expression in mature hippocampal neurons. (A)

Confocal section of a 28-DIV hippocampal neuron immunostained for KCC2. (Left) asterisk shows the soma. (Right) Enlarged boxed region shows intense immunoreactivity in spines (arrowheads). (Scale bar: Left, 5 μm; Right, 2 μm.) (B) Normalized fluorescence intensity per cluster of KCC2 staining in den-dritic shaft vs. spines in 30 cells from three independent cultures (***P < 0.001). (C) Effect of KCC2 silencing on protein expression in neurons at 24 DIV. Arrows and stars show dendrites and somata of transfected neurons, respectively. (Scale bar: 5 μm.) (D) Normalized fluorescence intensity per pixel of KCC2 staining in neurons expressing shNT (n = 10) or shKCC2 (n = 7) (P < 0.005). shKCC2, shRNA against KCC2; shNT, nontarget shRNA. (E) (Left) Confocal images of dendritic spines in neurons expressing shNT or shKCC2 showing GFP (green) and KCC2 (red) immunostaining. (Right) Line scans (from white dotted lines) of KCC2 immunofluorescence. The vertical axis shows raw KCC2 fluorescence intensity. (Scale bar: 0.5 μm.) (F) 3D recon-structions from tertiary dendrites of neurons expressing shNT or shKCC2 and GFP. (Scale bar: 5 μm.) (G) Quantification of spine length and head diameter from neurons expressing shNT or shKCC2.

Fig. 2. Decreased quantal size and GluR1 postsynaptic clustering upon sup-pression of KCC2. (A) (Left) Ten-second recordings of mEPSCs in neurons expressing nontarget shRNA or shRNA against KCC2. (Right) Scaled averages of 150 mEPSCs from the same recordings revealed no change in onset or decay kinetics. (B) Summary graphs of mEPSC amplitude distributions (Left), mean amplitude, and frequency (Right) in neurons expressing shNT (n = 27) or shKCC2 (n = 23) from five experiments. Suppression of KCC2 expression re-duced the amplitude of mEPSCs by ∼17% (P < 0.001 on distributions and P < 0.05 on means) without affecting their frequency (P = 0.2). (C) Representative sections of dendrites with GFP and GluR1 immunostaining. Arrowheads in-dicate dendritic spines. KCC2 suppression induces a reduction in GluR1 clusters in dendritic spines. (Scale bar: 1 μm.) (D) (Upper) Normalized fluorescence intensity per cluster of GluR1 staining (shNT, n = 55; shKCC2, n = 44) from three experiments (***P < 0.005). (Lower) Quantification of the proportion of spines bearing GluR1-immunopositive clusters from the same samples (P = 0.2).

P = 0.2) of mEPSCs (Fig. 3 A and B). Consistent with these observations, VU0240551 did not induce detectable changes in GluR1 cluster immunofluorescence in dendritic spines (+7.3 ± 8.9% of control; P = 0.2) or in the proportion of GluR1-immunopositive spines (84.1 ± 1.0% vs. 82.6 ± 1.2%; P = 0.7; Fig. 3 C and D). Therefore, decreased excitatory synapse efficacy upon suppression of KCC2 expression does not result from the loss of KCC2 function and subsequent reduction of GABA signaling.

The C-terminal domain of KCC2 interacts with the 4.1N protein (11), a neuronal FERM-domain protein (12) that binds both actin and the GluR1 subunit of AMPA receptors (23). Because stabilization of synaptic AMPA receptors depends critically on the actin cytoskeleton (24), direct or indirect teraction between AMPA receptors, actin, and KCC2 may in-fluence the maintenance of synaptic AMPA receptors. We therefore aimed at preventing KCC2-4.1N interaction without affecting KCC2 function by overexpressing the KCC2-CTD (11). We first verified that overexpressing KCC2-CTD did not affect KCC2 function by comparing the gradient in reversal potential of GABA currents (EGABA) induced at the soma vs. distal dendrites using local photolysis of caged GABA (11). In our conditions, the EGABA evoked by photolysis of RuBi-GABA (Ascent Sci-entific) onto the soma of hippocampal neurons was −51.3 ± 1.6 mV (n = 25; Fig. 4A), close to the equilibrium potential for Cl−

(−52.3 mV). In neurons expressing GFP only, we measured a gradient of 7.3 ± 1.6 mV per 100 μM between somatic and dendritic EGABA(Fig. 4 A and B). This gradient was not signif-icantly different in neurons overexpressing KCC2-CTD (7.2 ± 2.3 mV per 100 μM; P = 0.6). In contrast, it was strongly reduced in neurons expressing shRNA against KCC2 vs. nontarget shRNA (1.7 ± 1.1 mV vs. 6.6 ± 1.5 mV per 100 μM; P < 0.05) and in cells exposed to the KCC2 antagonist VU0240551 vs. DMSO alone (2.5 ± 0.5 mV vs. 6.4 ± 0.7 mV per 100 μM; P < 0.005). Therefore, overexpressing KCC2-CTD in hippocampal neurons does not affect their apparent chloride extrusion ca-pacity, suggesting that KCC2 transport is functional.

Overexpression of KCC2-CTD did not cause any significant change in the morphology of dendritic spines (Table S1) but decreased mEPSC amplitude to a similar extent as did sup-pression of KCC2 exsup-pression by RNAi (10.6 ± 0.3 pA vs. 13.1 ± 0.9 pA; P < 0.05; Fig. 4 C and D). The effect of KCC2-CTD was not associated with significant changes in mEPSC frequency (15.8 ± 1.6 Hz vs. 14.3 ± 3.0 Hz; P = 0.2) or rise time (0.88 ± 0.04 ms vs. 0.93 ± 0.0.3 ms; P = 0.1). Accordingly, GluR1 immunostaining showed reduced cluster intensity in the dendritic spines of neurons expressing KCC2-CTD vs. GFP only (−22.3 ± 8.1%; P < 0.005), whereas the proportion of GluR1-immuno-positive spines was unaffected (90.1 ± 1.6% vs. 85.7 ± 3.1%; P = 0.4; Fig. 4 E and F). Therefore, suppressing KCC2 expression or interfering with KCC2 interaction with cytoplasmic partners without affecting its transport function reduces the strength of excitatory synapses. This effect likely reflects reduced aggregation of GluR1-containing AMPA receptors in dendritic spines.

Fig. 3. Effect of KCC2 suppression is independent of KCC2 function. (A) Scaled averages of 100 mEPSCs from neurons treated for >72 h with either 6 μM VU0240551 in DMSO or DMSO alone. (B) Summary graphs of mEPSC amplitude (Left) and frequency (Right) in neurons treated with DMSO (n = 23) or VU0240551 (n = 24) from four experiments. KCC2 antagonist did not significantly affect mEPSC amplitude (P = 0.2) or frequency (P = 0.8). (C) GFP and GluR1 immunostaining of hippocampal neurons treated with either DMSO or VU0240551. Arrowheads indicate dendritic spines. (Scale bar: 1 μm.) (D) (Upper) Normalized fluorescence intensity per cluster of GluR1 staining in neurons treated with either DMSO (n = 44 cells) or VU0240551 (n = 40 cells) from three experiments (P = 0.2). (Lower) Quantification of the proportion of spines bearing GluR1-immunopositive clusters from the same samples (P = 0.7).

Fig. 4. Overexpression of KCC2-CTD mimics the effects of KCC2 suppression. (A) (Left) Currents evoked by local photolysis of RuBi-GABA on the soma or distal dendrite of somatically whole-cell patch-clamped neurons. (Right) Representative currents at voltage steps ranging from −95 to −35 mV, with corresponding normalized current/voltage relations. Note the leftward shift in current/voltage relations of dendritic vs. somatic GABA currents. (B) Summary data showing a similar shift in neurons expressing GFP vs. KCC2-CTD (P = 0.6) but not in neurons expressing KCC2-specific vs. nontarget shRNA (*P < 0.05) or in neurons treated with VU0240551 (6 μM) vs. DMSO alone (**P < 0.005). n = 10–12 cells for each condition. (C) Scaled averages of 100 mEPSCs recorded from neurons expressing either GFP or KCC2-CTD and GFP. (D) Mean mEPSC amplitude (Left) and frequency (Right) charts. Over-expression of KCC2-CTD significantly reduced mEPSC amplitude by ∼19% (*P < 0.05) but not frequency (P = 0.2). n = 22 KCC2-CTD cells and 20 GFP cells. (E) Representative sections of dendrites with GFP and GluR1 immunostaining. Arrowheads show dendritic spines. (Scale bar: 1 μm.) (F) Normalized fluo-rescence intensity per cluster of GluR1 staining (Upper, **P < 0.005) and proportion of GluR1-immunopositive spines (Lower, P = 0.4) in neurons expressing GFP (n = 28) or KCC2-CTD and GFP (n = 30) in two independent experiments.

Gauvain et al. PNAS Early Edition | 3 of 6

N E U R O S C IE N C E

Increased Lateral Diffusion of GluR1 Subunit in Dendritic Spines upon KCC2 Suppression.Synaptic AMPA receptor content is influenced by interactions with anchoring proteins, exocytosis, endocytosis, and lateral receptor diffusion in the plasma membrane (25, 26). KCC2 might then contribute to diffusional constraints for AMPA receptors in dendritic spines by interacting with 4.1N and actin cytoskeleton (17). We tested this hypothesis by monitoring AMPA receptor lateral diffusion with quantum dot (QD)-based single-particle tracking (27). Neurons were cotransfected with vectors expressing recombinant GluR1 GFP-tagged at its extra-cellular N terminus and either nontarget or KCC2-specific shRNA constructs. QD-bound recombinant GluR1 displayed heterogeneous diffusion behaviors in dendritic spines and shafts, as do native AMPA receptors (28). In dendritic spines, some receptors located in spine heads diffused slowly, exploring only a restricted membrane area (e.g., gray trajectory in Fig. 5B). Others, although confined to the spine head, diffused more rapidly over a larger membrane area (gray trajectory in Fig. 5C). These distinct behaviors may reflect differential synaptic an-choring or trapping of receptors in endocytic zones (28, 29). We therefore distinguished slowly (D # 1.5 × 10−2

$μm²$s⁻¹] and rapidly (D > 1.5 × 10−2

$μm²$s⁻¹) diffusing receptors, where D represents the diffusion coefficient. Consistent with the prefer-ential synaptic anchoring of GluR1 in dendritic spines (29), the proportion of slow QDs was larger in spines than on dendritic shafts (46.5 vs. 34.3%; n = 43 and n = 137 QDs, respectively).

KCC2 extinction did not affect the exploratory behavior or confinement of slow receptors in spines (Fig. 5 A and C). Their mean diffusion coefficient and confinement domain were not significantly different in neurons expressing KCC2-specific vs. nontarget shRNA (D = 0.82 ± 0.01 × 10−2

$μm²$s⁻¹vs. 0.85 ± 0.07 ×10−2

$μm²$s⁻¹, P = 0.6; L = 97 ± 6 nm vs. 111 ± 7 nm, P = 0.1), where L represents the confinement domain. In contrast, rapid QD-bound GluR1 in dendritic spines exhibited a twofold increase in diffusion coefficient and a 1.7-fold increase in con-finement domain in neurons expressing KCC2-specific vs. non-target shRNA (D = 12.05 ± 2.45 × 10−2

$μm²$s⁻¹ vs. 5.99 ± 1.40 × 10−2

$μm²$s⁻¹, P < 0.05; L = 611 ± 13 nm vs. 366 ± 5 nm, respectively, P < 0.05; Fig. 5 D and E). This effect was also reflected by the steeper slope of mean square displacement functions derived from trajectories in either condition (Fig. 5C). Thus, KCC2 apparently constrains, directly or indirectly, lateral movements of rapid GluR1-containing AMPA receptors in den-dritic spines. This effect seems to be independent of changes in spine head diameter, because control experiments showed that lateral diffusion of GluR1 was unrelated to spine size (Fig. S2). In contrast to its effect on GluR1 diffusion in spines, KCC2 sup-pression did not significantly alter diffusion on dendritic shafts. Thus, the mean diffusion coefficient and confinement domain of both slow and rapid QD-bound GluR1 were not significantly

dif-ferent in neurons expressing KCC2-specific vs. nontarget shRNA (Fig. S3). We conclude that KCC2 constrains GluR1 lateral dif-fusion specifically in dendritic spines but not on dendritic shafts. This constraint primarily affects rapidly diffusing GluR1, suggest-ing that distinct molecular interactions restrict diffusion of slow GluR1-containing AMPA receptors.

KCC2 Constrains Lateral Diffusion of Transmembrane but Not Membrane-Anchored Neuronal Cell Adhesion Molecule in Dendritic Spines.KCC2 molecules are densely clustered in dendritic spines (ref. 10 and our observations) and bind actin cytoskeleton through direct 4.1N interaction (11). They might then constrain lateral diffusion of membrane proteins in dendritic spines by specific cytoskeletal interactions or by nonspecific molecular crowding. To discriminate between these possibilities, we ex-amined the lateral diffusion of two distinct isoforms of the neuronal cell adhesion molecule (NCAM). NCAM 180 is a transmembrane isoform with a short intracellular domain that interacts with the actin cytoskeleton through β1-spectrin, whereas NCAM 120 lacks an intracellular domain and is mem-brane-anchored by a GPI (30) (Fig. 6A). We transfected neurons with chicken NCAM 180 or NCAM 120 together with nontarget or KCC2-specific shRNA. Consistent with their distinct mem-brane interactions, NCAM 180 explored a smaller area than NCAM 120 in dendritic spines (Fig. 6 B and C) and its diffusion in dendritic spines was 3.2-fold slower than that of NCAM 120 (Fig. 6D). In neurons in which KCC2 expression was suppressed by RNAi, membrane diffusion of NCAM 180 (black trajectory in Fig. 6B) but not NCAM 120 (black trajectory in Fig. 6C) was increased by almost twofold compared with neurons expressing nontarget shRNA (D = 6.62 ± 0.46 × 10−2

$μm²$s⁻¹vs. 3.81 ± 0.32 × 10−2

$μm²$s⁻¹; P < 0.001; Fig. 6D) and their confinement domain was increased by 1.5-fold (L = 440 ± 39 nm vs. 299 ± 18 nm; P < 0.003). Because the 180 isoform is the predominant NCAM in dendritic spines (31), we also examined endogenous NCAM in spines and observed a similar increase in lateral dif-fusion on suppression of KCC2 (Fig. S4). In contrast, NCAM 120 diffusion was not significantly affected in these conditions (D = 15.05 ± 1.78 × 10−2

$μm²$s⁻¹ vs. 12.18 ± 1.32 × 10−2

$μm²$s⁻¹, respectively; P = 0.2; Fig. 6D). These results show that KCC2 suppression specifically facilitates lateral diffusion of trans-membrane but not trans-membrane-anchored NCAM in dendritic spines. We conclude that KCC2 constrains protein diffusion in dendritic spines, likely via interactions with submembrane proteins